Blind source separation of nonlinearly mixed plant leaf electrical signals using polynomial-mapped FastICA

Area of Science:

• Plant electrophysiology
• Non-invasive sensing
• Signal processing

Background:

• Plant electrical signals arise from diverse cell types and exhibit nonlinear mixing.
• Traditional Independent Component Analysis (ICA) methods struggle with these nonlinearities, limiting analysis of non-invasive plant electrophysiological signals.
• Understanding these signals is crucial for advancing precision agriculture.

Purpose of the Study:

• To develop and validate a novel blind source separation (BSS) method for plant electrical signals with nonlinear characteristics.
• To overcome the limitations of conventional ICA in processing non-invasive plant electrophysiology.
• To improve the analysis of distinct cellular electrical activities in plants.

Main Methods:

• A FastICA algorithm combined with second-order polynomial mapping (Poly2) was developed for BSS.
• Surrogate data generated via phase randomization were used to confirm signal nonlinearity using sample entropy.
• Performance was evaluated against linear ICA, third-order polynomial ICA (Poly3), and B-spline algorithms using simulations and real plant data.

Main Results:

• Plant electrical signals were confirmed to possess intrinsic nonlinear dynamics.
• The Poly2 method significantly outperformed conventional linear ICA, Poly3, and B-spline algorithms in simulations.
• Poly2 achieved high Spearman correlation coefficients (S1: 0.82, S2: 0.87) in simulations and successfully separated components from real leaf surface recordings (S1: 0.80, S2: 0.82).

Conclusions:

• The proposed polynomial-mapped FastICA approach offers superior signal decoupling for plant electrophysiology.
• This method effectively distinguishes electrical signals from different cell types (e.g., guard and mesophyll cells).
• The findings advance BSS applications in precision agriculture and plant electrophysiology research.